Is any GTX 750 the GeForce GTX 750 Ti owner here?
 

I would like to test and to compare it with my GT 640 card in meter of cuda sieving.
So if is someone that have any of those cards pleae PM.
Thanks

hi, i'm sure TheJudger reacts soon.

GTX 750 is Maxwell series. So that's 512 cores. Or better 128 cores * 4 * 1.02Ghz
GTX 640 is Kepler series. 384 cores or 192 cores * 2 * 0.9Ghz

So it's 4 Streaming Multiprocessors (SM's) versus 2.

[url]http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#maximize-instruction-throughput[/url]

In 24 bits code TF (up to 72 bits) probably it's just Ghz * total number of cores that determines performance.

In 32 bits code TF (more than 72 bits) dominating factor will probably be integer Multiply-Add throughput. That's just 32 units each SM for the 640. So 64 for entire GPU, versus Maxwell will probably completely annihilate it there.

There are five versions of the GT 640. The plain OEM model is Fermi based. So it depends on which card it is exactly.

[url]https://en.m.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_600_Series[/url]

[QUOTE=Mark Rose;438312]There are five versions of the GT 640. The plain OEM model is Fermi based. So it depends on which card it is exactly.

[URL]https://en.m.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_600_Series[/URL][/QUOTE]

My GT640 have GK107 chip ( Fermi base) 384 CUDA cores, 1 GB DDR3 memory 901/893 MHz ( GPU/MEM)

[QUOTE=pepi37;438320]My GT640 have GK107 chip ( Fermi base) 384 CUDA cores, 1 GB DDR3 memory 901/893 MHz ( GPU/MEM)[/QUOTE]
"GK" would be a Kepler chip.

The attached image is the Fermi version, GF116 chip.

[QUOTE=kladner;438324]"GK" would be a Kepler chip.

The attached image is the Fermi version, GF116 chip.[/QUOTE]
Ok, so I have Kepler chip.
Will now, when we know what card I have somebody hear my call and send me PM.
I prepared short test, it take only few minutes.

On a 750Ti (5 SMs):
Exponent 79186511
GPUSievePrimes 80k

70-71: 159 Ghz-d/day
76-77: 158 Ghz-d/day

Is there something I am missing? Surely after changing kernels at 76 it should noticeably slow down.
It only seemed to run 32-bit kernels. I assume that is what is optimal on Maxwell. Is there an ini setting which changes this?

You might also want to look at a used GTX960 if it fits within your budget ($150).
1024 CUDA cores @1200MHz
But it does rquire a 6 or 8pin PEG connector....

[QUOTE=VictordeHolland;438340]You might also want to look at a used GTX960 if it fits within your budget ($150).
1024 CUDA cores @1200MHz
But it does rquire a 6 or 8pin PEG connector....[/QUOTE]
Before I buy anything I need to test it and see what is performance...

What specific test are you looking for results from, under what OS?

[QUOTE=VBCurtis;438342]What specific test are you looking for results from, under what OS?[/QUOTE]
I test under Windows using PPSieve cuda ( ver 0.2.3)
Since I get on my card 80Mp/s ( one instance) and 100Mp/s ( two instances) and since this card is in my range I can spend some money, I try to determine how fast is that card
Can can you test data if you wish: just start , wait few minutes to get stable speed and send me speed you get

It took 3 cores to max me out
3 cores: 67*3=201M p/sec
2 cores: 85*2=170M p/sec
1 core: 95M p/sec
Even then the card wasn't limited by TDP which was at about 50%. It wasn't willing to increase the voltage any further to allow the clock rate to be higher.

Thanks for reply and this info!